Opinion Spam Detection in Product Reviews Using Self-Training Semi-Supervised Learning Approach
Dini Adni Navastara, Ana Alimatus Zaqiyah, Chastine Fatichah · 2019
The review of a product can influence a buyer's decision to buy the product. In addition to influencing buyer decisions, fake reviews can also confuse buyers who are looking for product information from honest and genuine reviews. We need a system that can filter spam to reduce the negative influence on product selling and product review writings. Spam that will be detected is the type of brand only spam and not a review. Those types get the initial label through manual labeling. Manual labeling requires a lot of time and effort. Therefore, in this paper, we proposed a self-training semi-supervised learning approach. This method labels spam from the prediction of the labeled training data. The best results were obtained with a scenario without stemming, merging of review centric features and bigram, SMOTE borderline1 oversampling and Polynomial SVM kernel that has accuracy 86.33%.